Re: [OT] Weighted distribution of Numbers
By ear On Tue, Aug 6, 2019 at 07:51 Bob Sneidar via use-livecode < use-livecode@lists.runrev.com> wrote: > You mean to say relative math is easy? > > Bob S > > > > On Aug 5, 2019, at 14:57 , Stephen Barncard via use-livecode < > use-livecode@lists.runrev.com> wrote: > > > > This is the kind of math I use every day, without knowing what I’m doing. > > > > On Mon, Aug 5, 2019 at 09:21 Mark Wieder via use-livecode < > > use-livecode@lists.runrev.com> wrote: > > > >> On 8/5/19 9:00 AM, Mark Wieder via use-livecode wrote: > >>> have to weight the measured values to determine the maximum (and the Q > >>> as desired). > >> > >> Urk. Now it's my turn to have misspoken. > >> The maximum is easy to measure. > >> But looking at the clustering of values to determine the Q of the > >> bandpass filter requires a different calculation. > >> > >> -- > >> Mark Wieder > > ___ > use-livecode mailing list > use-livecode@lists.runrev.com > Please visit this url to subscribe, unsubscribe and manage your > subscription preferences: > http://lists.runrev.com/mailman/listinfo/use-livecode -- -- Stephen Barncard - Sebastopol Ca. USA - mixstream.org ___ use-livecode mailing list use-livecode@lists.runrev.com Please visit this url to subscribe, unsubscribe and manage your subscription preferences: http://lists.runrev.com/mailman/listinfo/use-livecode
Re: [OT] Weighted distribution of Numbers
You mean to say relative math is easy? Bob S > On Aug 5, 2019, at 14:57 , Stephen Barncard via use-livecode > wrote: > > This is the kind of math I use every day, without knowing what I’m doing. > > On Mon, Aug 5, 2019 at 09:21 Mark Wieder via use-livecode < > use-livecode@lists.runrev.com> wrote: > >> On 8/5/19 9:00 AM, Mark Wieder via use-livecode wrote: >>> have to weight the measured values to determine the maximum (and the Q >>> as desired). >> >> Urk. Now it's my turn to have misspoken. >> The maximum is easy to measure. >> But looking at the clustering of values to determine the Q of the >> bandpass filter requires a different calculation. >> >> -- >> Mark Wieder ___ use-livecode mailing list use-livecode@lists.runrev.com Please visit this url to subscribe, unsubscribe and manage your subscription preferences: http://lists.runrev.com/mailman/listinfo/use-livecode
Re: [OT] Weighted distribution of Numbers
This is the kind of math I use every day, without knowing what I’m doing. On Mon, Aug 5, 2019 at 09:21 Mark Wieder via use-livecode < use-livecode@lists.runrev.com> wrote: > On 8/5/19 9:00 AM, Mark Wieder via use-livecode wrote: > > have to weight the measured values to determine the maximum (and the Q > > as desired). > > Urk. Now it's my turn to have misspoken. > The maximum is easy to measure. > But looking at the clustering of values to determine the Q of the > bandpass filter requires a different calculation. > > -- > Mark Wieder > ahsoftw...@gmail.com > > ___ > use-livecode mailing list > use-livecode@lists.runrev.com > Please visit this url to subscribe, unsubscribe and manage your > subscription preferences: > http://lists.runrev.com/mailman/listinfo/use-livecode > -- -- Stephen Barncard - Sebastopol Ca. USA - mixstream.org ___ use-livecode mailing list use-livecode@lists.runrev.com Please visit this url to subscribe, unsubscribe and manage your subscription preferences: http://lists.runrev.com/mailman/listinfo/use-livecode
Re: [OT] Weighted distribution of Numbers
On 8/5/19 9:00 AM, Mark Wieder via use-livecode wrote: have to weight the measured values to determine the maximum (and the Q as desired). Urk. Now it's my turn to have misspoken. The maximum is easy to measure. But looking at the clustering of values to determine the Q of the bandpass filter requires a different calculation. -- Mark Wieder ahsoftw...@gmail.com ___ use-livecode mailing list use-livecode@lists.runrev.com Please visit this url to subscribe, unsubscribe and manage your subscription preferences: http://lists.runrev.com/mailman/listinfo/use-livecode
Re: [OT] Weighted distribution of Numbers
On 8/5/19 1:48 AM, hh via use-livecode wrote: [@Mark: A (weighted) mean is a location parameter, one number.] Yes, exactly. In sum, Dagobert wants to change the method on base of the raw data or change the raw data such that the results are the wished ones. (Honi soit qui mal y pense ...) Respectfully, I think you're looking at a different problem. Let's say that I have a bandpass filter with a range of 2kHz<->20kHz. If I'm just looking at the frequencies I would say the center is at ~10kHz. But since the parameters of the bandpass filter can be changed I would have to weight the measured values to determine the maximum (and the Q as desired). -- Mark Wieder ahsoftw...@gmail.com ___ use-livecode mailing list use-livecode@lists.runrev.com Please visit this url to subscribe, unsubscribe and manage your subscription preferences: http://lists.runrev.com/mailman/listinfo/use-livecode
Re: [OT] Weighted distribution of Numbers
When computing limits for distribution categories given frequencies the following may be useful: A number q is a p%-quantile of a data set If the percentage of data nums <= q is >= p% and the percentage of data nums >= q is >= (100-p)% For each percentage p there is an interval [lowerV,upperV] so that each number from that interval is p%-quantile of the data set. To make it unique some use in case lowerV < upperV the average of lowerV and upperV. -- d=data in lines, sorted ascending numeric -- p=percentage (num in range 0-100) function quantile p,d put the num of lines of d into N put N*p/100 into m0 put line ceil(N*p/100) of d into lowerV put line N+1 - ceil(N*(100-p)/100) of d into upperV -- return avg(lowerV,upperV) --> unique variant if lowerV=upperV then return lowerV else return lowerV,upperV end quantile For example quantile(50,d) returns the median of a data set, quantile(25,d), quantile(50,d), quantile(75,d) the quartiles. ___ use-livecode mailing list use-livecode@lists.runrev.com Please visit this url to subscribe, unsubscribe and manage your subscription preferences: http://lists.runrev.com/mailman/listinfo/use-livecode
Re: [OT] Weighted distribution of Numbers
> I wrote: > In order to find these limits simply sort the random data (a random > sample drawn out of the raw data) and take the values that have > approximately 30% or 80% of the values below them (no scaling needed > for that). In statistical terms: Find the 30% and 80% quantiles. Please forget my last correction (I'm tired), was: > Sorry, read here 70% instead of 80% to relate correctly to the example > 0-30 = bad, 31-70 = neutral, 71-100 = good. "In order ..." relates correctly to the example > ... expected to have frequencies of say 30%, 50%, 20%. ___ use-livecode mailing list use-livecode@lists.runrev.com Please visit this url to subscribe, unsubscribe and manage your subscription preferences: http://lists.runrev.com/mailman/listinfo/use-livecode
Re: [OT] Weighted distribution of Numbers
I wrote: > In order to find these limits simply sort the random data (a random > sample drawn out of the raw data) and take the values that have > approximately 30% or 80% of the values below them (no scaling needed > for that). In statistical terms: Find the 30% and 80% quantiles. Sorry, read here 70% instead of 80% to relate correctly to the example 0-30 = bad, 31-70 = neutral, 71-100 = good. ___ use-livecode mailing list use-livecode@lists.runrev.com Please visit this url to subscribe, unsubscribe and manage your subscription preferences: http://lists.runrev.com/mailman/listinfo/use-livecode
Re: [OT] Weighted distribution of Numbers
[@Mark: A (weighted) mean is a location parameter, one number.] Here the customer (say Dagobert Duck) wants to change/weight the distribution of the data. As Dar says, he could do a mapping from 0-800 to bins as "bad, neutral, good" simply by setting limits for the bins. For example 0-30 = bad, 31-70 = neutral, 71-100 = good. And make these limits transparent and show their frequencies as they are. But now Dagobert wants to "adjust" by (1) and/or (2): (1) Set the limits for the bins such that each bin has a relative frequency of 1/3 (or given relative frequencies). This is setting categories by their frequencies in order to interpret the frequency of the categories. (2) Change the raw data such that for the given limits each bin has a relative frequency of 1/3 (or given relative frequencies). This is filling categories by changing data in order to interpret the frequency of the categories of the changed data. In sum, Dagobert wants to change the method on base of the raw data or change the raw data such that the results are the wished ones. (Honi soit qui mal y pense ...) -- I would accept (1) if one argues from *theoretical* reasons that the bins are expected to have frequencies of say 30%, 50%, 20%. This could lead to limits on base of *some* (random part) of raw data: In order to find these limits simply sort the random data (a random sample drawn out of the raw data) and take the values that have approximately 30% or 80% of the values below them (no scaling needed for that). In statistical terms: Find the 30% and 80% quantiles. Then one could use these (transparent) limits for the *rest* of the raw data and new raw data and interpret the frequencies of the bins. ___ use-livecode mailing list use-livecode@lists.runrev.com Please visit this url to subscribe, unsubscribe and manage your subscription preferences: http://lists.runrev.com/mailman/listinfo/use-livecode
RE: [OT] Weighted distribution of Numbers
Thanks all! I will take a read on that link also. Ralph DiMola IT Director Evergreen Information Services rdim...@evergreeninfo.net -Original Message- From: use-livecode [mailto:use-livecode-boun...@lists.runrev.com] On Behalf Of Mark Wieder via use-livecode Sent: Sunday, August 04, 2019 7:07 PM To: dsc--- via use-livecode Cc: Mark Wieder Subject: Re: [OT] Weighted distribution of Numbers On 8/4/19 3:00 PM, dsc--- via use-livecode wrote: > I'm unsure how often 800 or so changes. I'll call it 800, it is just a name. Values can range from 0 through 800. > > You can map a number in that range to 0-1 by dividing by 800. That is, scaled1(n) is n/800. > > I guess you want to map each number n in that into one of 101 bins, 0 through 100. Yes - that's a weighted mean. Here's a simple explanation https://sciencing.com/calculate-weighted-average-5328019.html ...and no, Hermann, it's (you'd think I'd know better than to argue with a real mathematician here, but...) not lying, it's the addition of another variable. -- Mark Wieder ahsoftw...@gmail.com ___ use-livecode mailing list use-livecode@lists.runrev.com Please visit this url to subscribe, unsubscribe and manage your subscription preferences: http://lists.runrev.com/mailman/listinfo/use-livecode ___ use-livecode mailing list use-livecode@lists.runrev.com Please visit this url to subscribe, unsubscribe and manage your subscription preferences: http://lists.runrev.com/mailman/listinfo/use-livecode
Re: [OT] Weighted distribution of Numbers
On 8/4/19 3:00 PM, dsc--- via use-livecode wrote: I'm unsure how often 800 or so changes. I'll call it 800, it is just a name. Values can range from 0 through 800. You can map a number in that range to 0-1 by dividing by 800. That is, scaled1(n) is n/800. I guess you want to map each number n in that into one of 101 bins, 0 through 100. Yes - that's a weighted mean. Here's a simple explanation https://sciencing.com/calculate-weighted-average-5328019.html ...and no, Hermann, it's (you'd think I'd know better than to argue with a real mathematician here, but...) not lying, it's the addition of another variable. -- Mark Wieder ahsoftw...@gmail.com ___ use-livecode mailing list use-livecode@lists.runrev.com Please visit this url to subscribe, unsubscribe and manage your subscription preferences: http://lists.runrev.com/mailman/listinfo/use-livecode
Re: [OT] Weighted distribution of Numbers
I'm unsure how often 800 or so changes. I'll call it 800, it is just a name. Values can range from 0 through 800. You can map a number in that range to 0-1 by dividing by 800. That is, scaled1(n) is n/800. I guess you want to map each number n in that into one of 101 bins, 0 through 100. Perhaps something like this: trunc( 100.9 * scaled(n) ) However you want to tweak at the top, so this might become trunc( 100.9 * tweak( scaled(n) ) ) The function tweak takes a number from 0 to 1 and returns a number from 0 to 1 where tweak(0) = 0 tweak(1) = 1 if x >= y then tweak(x) >= tweak(y) A straight line fits that but that isn't what you want. Here are some definitions for tweak() you might try: x [that straight line] 2*x/(1+x) sin(x * pi/2) some other sine thing that is symmetrical piecewise linear 1- sqrt( 1-x ) min( 1, 1.03*x) [special case of piecewise] However, if you want tweaking to depend on the entire dataset, then there is more work. > On Aug 4, 2019, at 3:09 PM, Dar Scott Consulting via use-livecode > wrote: > > Oh, good. I was worrying that you might have a bad customer. > >> On Aug 4, 2019, at 3:05 PM, Ralph DiMola via use-livecode >> wrote: >> >> >> I'm not plotting this but using it for searching. >> >> I'm not really lying. I'm trying to come up with the raw numbers from many >> individual components. It's like "Gone with the Wind" and "Apocalypse Now" >> both getting 100 on Rotten Tomatoes. But if you looked under the hood and >> added up components such as sound, costumes, artwork, casting... and applied >> a weight to each then "Apocalypse Now" might get a raw rating of 800 and >> "Gone with the Wind" get a 790. But they are both so close to the top I >> would want them to both get 100. I can do this via the "human factor" by >> manually adjusting some of the results(mostly at the top) but I would like >> to somewhat automate it so when the components change I will do a >> re-calculation run and say the top number goes up by 25 all the manual >> adjustments go out the window. I want this to be somewhat automated. >> >> Ralph DiMola >> IT Director >> Evergreen Information Services >> rdim...@evergreeninfo.net >> >> >> -Original Message- >> From: use-livecode [mailto:use-livecode-boun...@lists.runrev.com] On Behalf >> Of Dar Scott Consulting via use-livecode >> Sent: Sunday, August 04, 2019 4:33 PM >> To: How to use LiveCode >> Cc: Dar Scott Consulting >> Subject: Re: [OT] Weighted distribution of Numbers >> >> I was thinking the same, but was to afraid to say it. Yes, the actual name >> is "lying". >> >> However, there might be an honest attempt to display crowded dots or icons. >> >>> On Aug 4, 2019, at 2:19 PM, hh via use-livecode >> wrote: >>> >>>> Ralph D. wrote: >>>> I'm sure there's an actual name for doing this in the statistician's >>>> world but I don't know what it is. >>> >>> This has nothing to do with "statistics". >>> This is simply "try to lie by data cheating". >>> >>> >>> ___ >>> use-livecode mailing list >>> use-livecode@lists.runrev.com >>> Please visit this url to subscribe, unsubscribe and manage your >> subscription preferences: >>> http://lists.runrev.com/mailman/listinfo/use-livecode >> >> >> ___ >> use-livecode mailing list >> use-livecode@lists.runrev.com >> Please visit this url to subscribe, unsubscribe and manage your subscription >> preferences: >> http://lists.runrev.com/mailman/listinfo/use-livecode >> >> >> ___ >> use-livecode mailing list >> use-livecode@lists.runrev.com >> Please visit this url to subscribe, unsubscribe and manage your subscription >> preferences: >> http://lists.runrev.com/mailman/listinfo/use-livecode > > > ___ > use-livecode mailing list > use-livecode@lists.runrev.com > Please visit this url to subscribe, unsubscribe and manage your subscription > preferences: > http://lists.runrev.com/mailman/listinfo/use-livecode ___ use-livecode mailing list use-livecode@lists.runrev.com Please visit this url to subscribe, unsubscribe and manage your subscription preferences: http://lists.runrev.com/mailman/listinfo/use-livecode
Re: [OT] Weighted distribution of Numbers
Oh, good. I was worrying that you might have a bad customer. > On Aug 4, 2019, at 3:05 PM, Ralph DiMola via use-livecode > wrote: > > > I'm not plotting this but using it for searching. > > I'm not really lying. I'm trying to come up with the raw numbers from many > individual components. It's like "Gone with the Wind" and "Apocalypse Now" > both getting 100 on Rotten Tomatoes. But if you looked under the hood and > added up components such as sound, costumes, artwork, casting... and applied > a weight to each then "Apocalypse Now" might get a raw rating of 800 and > "Gone with the Wind" get a 790. But they are both so close to the top I > would want them to both get 100. I can do this via the "human factor" by > manually adjusting some of the results(mostly at the top) but I would like > to somewhat automate it so when the components change I will do a > re-calculation run and say the top number goes up by 25 all the manual > adjustments go out the window. I want this to be somewhat automated. > > Ralph DiMola > IT Director > Evergreen Information Services > rdim...@evergreeninfo.net > > > -Original Message- > From: use-livecode [mailto:use-livecode-boun...@lists.runrev.com] On Behalf > Of Dar Scott Consulting via use-livecode > Sent: Sunday, August 04, 2019 4:33 PM > To: How to use LiveCode > Cc: Dar Scott Consulting > Subject: Re: [OT] Weighted distribution of Numbers > > I was thinking the same, but was to afraid to say it. Yes, the actual name > is "lying". > > However, there might be an honest attempt to display crowded dots or icons. > >> On Aug 4, 2019, at 2:19 PM, hh via use-livecode > wrote: >> >>> Ralph D. wrote: >>> I'm sure there's an actual name for doing this in the statistician's >>> world but I don't know what it is. >> >> This has nothing to do with "statistics". >> This is simply "try to lie by data cheating". >> >> >> ___ >> use-livecode mailing list >> use-livecode@lists.runrev.com >> Please visit this url to subscribe, unsubscribe and manage your > subscription preferences: >> http://lists.runrev.com/mailman/listinfo/use-livecode > > > ___ > use-livecode mailing list > use-livecode@lists.runrev.com > Please visit this url to subscribe, unsubscribe and manage your subscription > preferences: > http://lists.runrev.com/mailman/listinfo/use-livecode > > > ___ > use-livecode mailing list > use-livecode@lists.runrev.com > Please visit this url to subscribe, unsubscribe and manage your subscription > preferences: > http://lists.runrev.com/mailman/listinfo/use-livecode ___ use-livecode mailing list use-livecode@lists.runrev.com Please visit this url to subscribe, unsubscribe and manage your subscription preferences: http://lists.runrev.com/mailman/listinfo/use-livecode
RE: [OT] Weighted distribution of Numbers
I'm not plotting this but using it for searching. I'm not really lying. I'm trying to come up with the raw numbers from many individual components. It's like "Gone with the Wind" and "Apocalypse Now" both getting 100 on Rotten Tomatoes. But if you looked under the hood and added up components such as sound, costumes, artwork, casting... and applied a weight to each then "Apocalypse Now" might get a raw rating of 800 and "Gone with the Wind" get a 790. But they are both so close to the top I would want them to both get 100. I can do this via the "human factor" by manually adjusting some of the results(mostly at the top) but I would like to somewhat automate it so when the components change I will do a re-calculation run and say the top number goes up by 25 all the manual adjustments go out the window. I want this to be somewhat automated. Ralph DiMola IT Director Evergreen Information Services rdim...@evergreeninfo.net -Original Message- From: use-livecode [mailto:use-livecode-boun...@lists.runrev.com] On Behalf Of Dar Scott Consulting via use-livecode Sent: Sunday, August 04, 2019 4:33 PM To: How to use LiveCode Cc: Dar Scott Consulting Subject: Re: [OT] Weighted distribution of Numbers I was thinking the same, but was to afraid to say it. Yes, the actual name is "lying". However, there might be an honest attempt to display crowded dots or icons. > On Aug 4, 2019, at 2:19 PM, hh via use-livecode wrote: > >> Ralph D. wrote: >> I'm sure there's an actual name for doing this in the statistician's >> world but I don't know what it is. > > This has nothing to do with "statistics". > This is simply "try to lie by data cheating". > > > ___ > use-livecode mailing list > use-livecode@lists.runrev.com > Please visit this url to subscribe, unsubscribe and manage your subscription preferences: > http://lists.runrev.com/mailman/listinfo/use-livecode ___ use-livecode mailing list use-livecode@lists.runrev.com Please visit this url to subscribe, unsubscribe and manage your subscription preferences: http://lists.runrev.com/mailman/listinfo/use-livecode ___ use-livecode mailing list use-livecode@lists.runrev.com Please visit this url to subscribe, unsubscribe and manage your subscription preferences: http://lists.runrev.com/mailman/listinfo/use-livecode
Re: [OT] Weighted distribution of Numbers
I was thinking the same, but was to afraid to say it. Yes, the actual name is "lying". However, there might be an honest attempt to display crowded dots or icons. > On Aug 4, 2019, at 2:19 PM, hh via use-livecode > wrote: > >> Ralph D. wrote: >> I'm sure there's an actual name for doing this in the statistician's >> world but I don't know what it is. > > This has nothing to do with "statistics". > This is simply "try to lie by data cheating". > > > ___ > use-livecode mailing list > use-livecode@lists.runrev.com > Please visit this url to subscribe, unsubscribe and manage your subscription > preferences: > http://lists.runrev.com/mailman/listinfo/use-livecode ___ use-livecode mailing list use-livecode@lists.runrev.com Please visit this url to subscribe, unsubscribe and manage your subscription preferences: http://lists.runrev.com/mailman/listinfo/use-livecode
Re: [OT] Weighted distribution of Numbers
Perhaps what you want is histogram smoothing or histogram curve fitting. Is this for a dot or icon display? Or for a plotted curve? > On Aug 4, 2019, at 1:38 PM, Ralph DiMola via use-livecode > wrote: > > Dar, > > Thanks for looking at this... > > These numbers are quality ratings. The raw numbers range from 0 to a max of > 800 or so. The customer wants to see a rating from 0-100 so I normalize them > into a range of 0 to 100 where the raw 0 is 0 and the raw 800 is 100. This > works perfectly. When looking at the resulting 0-100 ratings is where they > see the distribution anomalies. They would like to see the top numbers(say > from 94 to 100) to go to 100 and then the original 93 to be 99 and the > original 90 to be 97 or so. And also smooth out any gaps in the distribution > so there for example if there are almost no numbers in the 40s to bump up > the 30s a little and bump down the 50s a little. I'm sure there's an actual > name for doing this in the statistician's world but I don't know what it is. > > Ralph DiMola > IT Director > Evergreen Information Services > rdim...@evergreeninfo.net > Phone: 518-636-3998 Ex:11 > Cell: 518-796-9332 > > > -Original Message- > From: use-livecode [mailto:use-livecode-boun...@lists.runrev.com] On Behalf > Of Dar Scott Consulting via use-livecode > Sent: Sunday, August 04, 2019 3:03 PM > To: How to use LiveCode > Cc: Dar Scott Consulting > Subject: Re: [OT] Weighted distribution of Numbers > > Just to clarify... Is this right? > > The max of the raw numbers maps to 100. > The min of the raw numbers maps to 0. (Or is it 0 maps to 0?) The middle > number maps to something like 70. (Or is it half of the max maps to 70?) The > mapping is smooth. > > Where 70 might be something else. > >> On Aug 4, 2019, at 12:49 PM, Ralph DiMola via use-livecode > wrote: >> >> I have a set of raw numbers(6,000 of them from 0 to 800 or so). It was >> easy to normalize these numbers from 0 to 100. But as I look at the >> results I see that there is one at to top(100) and a few in the 90s >> and many more in the 70s and 80s. I need to make these numbers more >> evenly distributed and weighted towards the top(so the top few are >> 100) based on the current distribution of the raw numbers. I'm not a >> math whiz and not afraid to admit that going beyond linier equations >> is way over my head. From some searches I see the some sort of >> nonlinear regression is in order(I think)? Or a apply a log (like an >> audio log taper of a potentiometer)? I don't know... Can anyone point me > in the in the right direction? >> >> Thanks! >> >> Ralph DiMola >> IT Director >> Evergreen Information Services >> rdim...@evergreeninfo.net >> >> >> ___ >> use-livecode mailing list >> use-livecode@lists.runrev.com >> Please visit this url to subscribe, unsubscribe and manage your > subscription preferences: >> http://lists.runrev.com/mailman/listinfo/use-livecode > > > ___ > use-livecode mailing list > use-livecode@lists.runrev.com > Please visit this url to subscribe, unsubscribe and manage your subscription > preferences: > http://lists.runrev.com/mailman/listinfo/use-livecode > > > ___ > use-livecode mailing list > use-livecode@lists.runrev.com > Please visit this url to subscribe, unsubscribe and manage your subscription > preferences: > http://lists.runrev.com/mailman/listinfo/use-livecode ___ use-livecode mailing list use-livecode@lists.runrev.com Please visit this url to subscribe, unsubscribe and manage your subscription preferences: http://lists.runrev.com/mailman/listinfo/use-livecode
Re: [OT] Weighted distribution of Numbers
> Ralph D. wrote: > I'm sure there's an actual name for doing this in the statistician's > world but I don't know what it is. This has nothing to do with "statistics". This is simply "try to lie by data cheating". ___ use-livecode mailing list use-livecode@lists.runrev.com Please visit this url to subscribe, unsubscribe and manage your subscription preferences: http://lists.runrev.com/mailman/listinfo/use-livecode
RE: [OT] Weighted distribution of Numbers
Dar, Thanks for looking at this... These numbers are quality ratings. The raw numbers range from 0 to a max of 800 or so. The customer wants to see a rating from 0-100 so I normalize them into a range of 0 to 100 where the raw 0 is 0 and the raw 800 is 100. This works perfectly. When looking at the resulting 0-100 ratings is where they see the distribution anomalies. They would like to see the top numbers(say from 94 to 100) to go to 100 and then the original 93 to be 99 and the original 90 to be 97 or so. And also smooth out any gaps in the distribution so there for example if there are almost no numbers in the 40s to bump up the 30s a little and bump down the 50s a little. I'm sure there's an actual name for doing this in the statistician's world but I don't know what it is. Ralph DiMola IT Director Evergreen Information Services rdim...@evergreeninfo.net Phone: 518-636-3998 Ex:11 Cell: 518-796-9332 -Original Message- From: use-livecode [mailto:use-livecode-boun...@lists.runrev.com] On Behalf Of Dar Scott Consulting via use-livecode Sent: Sunday, August 04, 2019 3:03 PM To: How to use LiveCode Cc: Dar Scott Consulting Subject: Re: [OT] Weighted distribution of Numbers Just to clarify... Is this right? The max of the raw numbers maps to 100. The min of the raw numbers maps to 0. (Or is it 0 maps to 0?) The middle number maps to something like 70. (Or is it half of the max maps to 70?) The mapping is smooth. Where 70 might be something else. > On Aug 4, 2019, at 12:49 PM, Ralph DiMola via use-livecode wrote: > > I have a set of raw numbers(6,000 of them from 0 to 800 or so). It was > easy to normalize these numbers from 0 to 100. But as I look at the > results I see that there is one at to top(100) and a few in the 90s > and many more in the 70s and 80s. I need to make these numbers more > evenly distributed and weighted towards the top(so the top few are > 100) based on the current distribution of the raw numbers. I'm not a > math whiz and not afraid to admit that going beyond linier equations > is way over my head. From some searches I see the some sort of > nonlinear regression is in order(I think)? Or a apply a log (like an > audio log taper of a potentiometer)? I don't know... Can anyone point me in the in the right direction? > > Thanks! > > Ralph DiMola > IT Director > Evergreen Information Services > rdim...@evergreeninfo.net > > > ___ > use-livecode mailing list > use-livecode@lists.runrev.com > Please visit this url to subscribe, unsubscribe and manage your subscription preferences: > http://lists.runrev.com/mailman/listinfo/use-livecode ___ use-livecode mailing list use-livecode@lists.runrev.com Please visit this url to subscribe, unsubscribe and manage your subscription preferences: http://lists.runrev.com/mailman/listinfo/use-livecode ___ use-livecode mailing list use-livecode@lists.runrev.com Please visit this url to subscribe, unsubscribe and manage your subscription preferences: http://lists.runrev.com/mailman/listinfo/use-livecode
Re: [OT] Weighted distribution of Numbers
On 8/4/19 11:49 AM, Ralph DiMola via use-livecode wrote: I have a set of raw numbers(6,000 of them from 0 to 800 or so). It was easy to normalize these numbers from 0 to 100. But as I look at the results I see that there is one at to top(100) and a few in the 90s and many more in the 70s and 80s. I need to make these numbers more evenly distributed and weighted towards the top(so the top few are 100) based on the current distribution of the raw numbers. I'm not a math whiz and not afraid to admit that going beyond linier equations is way over my head. From some searches I see the some sort of nonlinear regression is in order(I think)? Or a apply a log (like an audio log taper of a potentiometer)? I don't know... Can anyone point me in the in the right direction? Someone will no doubt correct me on this, but it sounds like you want the weighted mean of the data set. Something like repeat for each value in the list add (the value / the number of values) to tWeightedMean end repeat -- Mark Wieder ahsoftw...@gmail.com ___ use-livecode mailing list use-livecode@lists.runrev.com Please visit this url to subscribe, unsubscribe and manage your subscription preferences: http://lists.runrev.com/mailman/listinfo/use-livecode
Re: [OT] Weighted distribution of Numbers
Just to clarify... Is this right? The max of the raw numbers maps to 100. The min of the raw numbers maps to 0. (Or is it 0 maps to 0?) The middle number maps to something like 70. (Or is it half of the max maps to 70?) The mapping is smooth. Where 70 might be something else. > On Aug 4, 2019, at 12:49 PM, Ralph DiMola via use-livecode > wrote: > > I have a set of raw numbers(6,000 of them from 0 to 800 or so). It was easy > to normalize these numbers from 0 to 100. But as I look at the results I see > that there is one at to top(100) and a few in the 90s and many more in the > 70s and 80s. I need to make these numbers more evenly distributed and > weighted towards the top(so the top few are 100) based on the current > distribution of the raw numbers. I'm not a math whiz and not afraid to admit > that going beyond linier equations is way over my head. From some searches I > see the some sort of nonlinear regression is in order(I think)? Or a apply a > log (like an audio log taper of a potentiometer)? I don't know... Can anyone > point me in the in the right direction? > > Thanks! > > Ralph DiMola > IT Director > Evergreen Information Services > rdim...@evergreeninfo.net > > > ___ > use-livecode mailing list > use-livecode@lists.runrev.com > Please visit this url to subscribe, unsubscribe and manage your subscription > preferences: > http://lists.runrev.com/mailman/listinfo/use-livecode ___ use-livecode mailing list use-livecode@lists.runrev.com Please visit this url to subscribe, unsubscribe and manage your subscription preferences: http://lists.runrev.com/mailman/listinfo/use-livecode
